Productside Stories

How AI Is Changing Product Management

Featured Guest:

Advaita Nigudkar | Director of Product Management at BILL
09/15/2026

Summary

In this episode of Productside Stories, Rina Alexin sits down with Advaita Nigudkar, Director of Product Management at BILL, to talk about what changes and what stubbornly does not when AI enters the product development lifecycle.

Advaita spent eight years at BILL building platform-level products and led the zero-to-one launch of the company’s first agentic AI workflow. She shares the context engineering work her team has built to make AI output usable at scale: skills that interrogate the PM before a single word gets drafted, best past PRDs handed to a language model as the gold standard, and the trust features designed for a platform that moves real money for small businesses.

The conversation covers how AI is freeing up time that should go into customer interviews and product teardowns, not more documentation, and what it takes to hold a quality bar as teams grow and churn. Advaita closes with the warning she gives every PM who gets comfortable with easy answers: when curiosity disappears, so does the thing that made you good at this job.

Key Takeaways

Garbage in, garbage out is now a job description.

  • One-line prompts produce PRDs that solve the admin use case and ignore everyone else. Advaita’s team built skills that interrogate the PM first: which customer, which SKU, which partner type. Better context is not optional, it is the work.

An LLM as your PRD judge.

  • Feed the model your best past PRDs as the gold standard and let it flag what is missing. The PM supplies the substance. The model polishes. Consistency survives headcount growth and churn because the standard lives in the system, not in one person’s head.

The trust feature you hope nobody opens.

  • BILL’s agents move money, so the team shipped an activity log and on/off controls. Engagement dropped once customers trusted it. That was the point. Auditability is not a feature customers should need constantly. It is the feature that makes everything else feel safe.

AI gives back time. Use it for discovery, not more pages.

  • The time AI returns belongs in customer interviews and product teardowns, not in generating more documents nobody reads carefully. Advaita’s team is expected to spend the freed capacity going deeper, not faster.

Curiosity is what AI can quietly take from you.

  • When answers come easily, you stop asking. That is the real risk. Advaita’s warning to every PM: do not let easy outputs replace the habit of sitting with a question, watching a recording, or going deeper than the model’s first draft.

Chapters

  • 00:00:00 – Introduction and Guest Background
  • 00:02:16 – AI’s Influence on Product Craft Pillars
  • 00:03:11 – The Role of Human Judgment Versus AI in Product Differentiation
  • 00:05:09 – Challenges of Cross-Functional Collaboration with AI
  • 00:06:14 – Using AI to Represent Stakeholders and Junior PMs
  • 00:07:22 – Evolution of Product Management Roles Over Five Years
  • 00:08:48 – Impact of AI Since 2022 on Productivity
  • 00:10:26 – AI’s Role in Managing Platform Complexity
  • 00:11:52 – Improving AI Prompt Engineering and Output Quality
  • 00:13:20 – Using LLMs as Judges for Product Documentation
  • 00:15:39 – Maintaining High Standards with AI-Generated Work
  • 00:17:28 – Enabling Teams with Standardized AI Prompts
  • 00:20:06 – Building AI Skills for Company-Wide Workflows
  • 00:21:09 – Building Trust in AI in Regulated Environments
  • 00:23:02 – Transparency Features for AI Products
  • 00:25:52 – Future Use Cases for AI in Product Management
  • 00:27:22 – Gaining More Strategic Time with AI
  • 00:28:42 – Advice for Product Managers on AI Adoption
  • 00:30:14 – The Importance of Curiosity in AI-Driven Product Management
  • 00:30:55 – Connecting with Advaita and Closing Remarks

Why Listen to This Episode?

In this episode, you’ll get:

  • A repeatable way to raise the bar on AI-written PRDs, using past work as the standard instead of vibes
  • The context you have to supply before AI is useful: personas, permissions, business types, historical decisions
  • A trust playbook for AI features in regulated environments, from encrypted data handling to auditability and kill switches
  • A framework for what stays human: negotiation, cross-functional trade-offs, and the judgment behind which problems deserve solving
  • Plus the warning Advaita gives every PM who gets comfortable with easy answers

 

Introduction

Rina Alexin | 00:00:00 – 00:02:16

Hi everyone and welcome to Productside Stories, the podcast where we dig into the very real and raw lessons learned from product leaders and thinkers all over the world. I’m your host, Rina Alexin, CEO of Productside, and today I’m talking with Advaita Nigudkar, Director of Product Management at BILL, who worked on the zero-to-one launch of the company’s first AI agentic platform. We’ll be talking about how AI is reshaping the day-to-day of product management, which I think is a topic on everybody’s mind. Welcome, Advaita.

Advaita Nigudkar | 00:00:37

Thank you. I’m happy to be here.

Rina Alexin | 00:00:40

I always love to hear everyone’s stories. Here’s my first question for you: how did you personally end up in product management?

Advaita Nigudkar | 00:00:49 – 00:02:16

I think I’ve always been a very curious kid, and as I grew up I found out about product management as a role. I was an IT engineer and I wanted to try my hand at it. I figured I’d give it a shot, and if I didn’t like it I’d go back to coding. I tried it and it has been eight years at BILL with never a single look back. It’s a domain that gives me immense joy, whether it’s peeling the onion on customer problems, coming up with solutions, working with cross-functional teams, or shipping quality products to our end customers.

Rina Alexin | 01:25

It’s good that you started out with curiosity, because I still hear it’s probably the number one thing people are trying to hire for in great product managers. In your early career, was there a product principle that really stuck with you?

Advaita Nigudkar | 01:43 – 00:02:16

The one that comes to mind is customer obsession. Not in the poster-on-the-wall kind of way, but how do you truly understand what problem you’re solving for so that you come up with the right solutions? Especially in the world of AI, we’re all eager to build and ship solutions. But if you get the problem wrong, the solutions you build will always be wrong. They won’t actually address the customer’s pain point. That’s the principle I believe in.

AI’s Influence on Product Craft Pillars

Rina Alexin | 00:02:16 – 00:03:11

The reason I asked that question is that even though AI is definitely changing a lot about product management, it hasn’t really changed whether we need to be customer-obsessed or problem-focused. Has it?

Advaita Nigudkar | 00:02:34 – 00:03:11

It’s actually even more important in today’s world. Everyone is eager to go solve the next big problem or come up with the next fancy solution, but that’s not always in tune with what customer problems we’re actually dealing with. Once you get the customer problem right, your solution almost always follows.

Rina Alexin | 00:03:01

That’s exactly where I want to start this conversation: even if AI is changing a lot, what isn’t changing about our roles today?

Advaita Nigudkar | 00:03:14 – 00:03:11

Product craft, full stop. As product managers, you need to be really good at your craft, and it comes up in three main pillars:

  • Strategy: the qualitative aspect, talking to customers, understanding and observing their day-to-day with or without the product or the problem you’re solving for
  • Problem and solution shaping: defining what the actual opportunity is before committing to a direction
  • Metrics: knowing how you’ll measure whether the solution is working

AI can help in each of these pillars, but your context of the problem, of the space you’re in, is going to continue to be irreplaceable. The higher the quality of context you can offer, the better your outputs with AI are going to be.

The Role of Human Judgment Versus AI in Product Differentiation

Rina Alexin | 00:03:11 – 00:05:09

One of the things that comes up constantly in my conversations with product leaders is that product management is really about judging what to build before you build it. And if you start to delegate that conceptual thinking to AI, isn’t there a danger of sameness? The differentiation between products happens because humans are influencing the decision to lean in one direction or another. If companies shift to thinking they can just AI their way through that context-gathering and thinking, is there a risk that products lose differentiation?

Advaita Nigudkar | 00:05:12 – 00:05:09

Yes, definitely. If you think about what AI does really well right now, it does a very good job when you tell it what tasks to do. But if you take a step back and think about the PM’s role, a big part of it is working cross-functionally: the negotiation, the prioritization, the trade-off discussions. I don’t believe we’re at a stage yet where you can hand those human elements to AI and say: go talk to risk, legal, and compliance, work with these teams, figure out what trade-off makes sense. That’s a lot of internal context that PMs bring in. Cross-functional collaboration is still where PMs create irreplaceable value.

Challenges of Cross-Functional Collaboration with AI

Rina Alexin | 00:05:09 – 00:06:14

I have heard of product managers creating AI agents to represent their stakeholders. Have you seen that in practice?

Advaita Nigudkar | 00:06:17 – 00:06:14

I’ve heard of it but haven’t seen it in practice, so I’m genuinely curious what the results have been like.

Rina Alexin | 00:06:26

I can imagine it helps somewhat, especially for junior product managers. It’s a way to have those challenging discussions with an AI stand-in for a sales leader before you go talk to the real person who needs to make quota this month. Good preparation, even if it isn’t the same as the real conversation.

Using AI to Represent Stakeholders and Junior PMs

Rina Alexin | 00:06:14 – 00:07:22

So we’ve talked about ways AI maybe isn’t changing the role. Now flip it: think about your own role. What makes it unrecognizable today compared to five years ago?

Advaita Nigudkar | 00:07:00 – 00:07:22

Five years ago it was a lot more traditional product management. For me personally, my journey has also evolved beyond AI: I used to be a vertical PM, owning a slice of the product, building and shipping features to end customers. Then I got the opportunity to work on a unified platform initiative, which was about thinking horizontally and figuring out how to build once and scale to five or six verticals. That switched something in my head: this is how you think when you’re building for scale. How do you build it once so you can repurpose most pieces and move faster as teams onboard?

Since then I’ve ended up in a more platform-oriented space, where every time I try to build something I first ask: how can we do this and serve multiple consumers out of one solution?

Evolution of Product Management Roles Over Five Years

Rina Alexin | 00:07:22 – 00:08:48

And going back to the before-and-after with AI specifically?

Advaita Nigudkar | 00:08:00 – 00:08:48

Starting around 2022 is when we all had access to these LLMs, and it’s changed my professional life. Whether it’s data analysis, UX prototypes, quickly riffing with AI to sanity-check a thesis or get a counter-argument, the time to do all of that has reduced significantly. It opens up more time for deeper analysis or more strategic work. And today almost everyone logs on, opens their AI tool, and starts their day there. That wasn’t a pattern before. Personally, if I have a question, I no longer go to Google. I just open Claude or ChatGPT, get my question addressed, and move on. It’s a paradigm shift.

Impact of AI Since 2022 on Productivity

Rina Alexin | 00:08:48 – 00:10:26

Platform is probably the hardest product to own. In B2C you have direct access to your user. In B2B you’re separated from your user. But in platform, the complexity of the number of personas you have to keep in mind when making decisions makes it genuinely hard. Have you found AI helps with that complexity?

Advaita Nigudkar | 00:10:35 – 00:10:26

It’s started to help, but it’s required a lot of context I’ve had to supply. AI can draw on its broader knowledge base, but what I’ve started doing is creating templates that describe the different kinds of companies, roles, permissions, personas, and business types relevant to our platform. It’s unfair to expect AI to have all the historical context we’ve accumulated, but that’s where training comes in: this is the reference, this is the context, these are the use cases to think of when helping me with a solution.

When I used to give it a prompt without that context, it would solve for the admin use case, the most common one, and miss everyone else. The question is always: how do you make sure your solution is holistic and not just solving for one apple on the tree?

AI’s Role in Managing Platform Complexity

Rina Alexin | 00:10:26 – 00:11:52

What have you done differently to get better output? Is it just better prompting over time, or more specific types of context?

Advaita Nigudkar | 00:12:00 – 00:11:52

It’s been a journey. Garbage in, garbage out: if I give a single prompt saying “write me a PRD for adding users,” it will do that, but it won’t understand the context of the universe you’re in, and it won’t give you a solution that works. There’s so much back and forth. The goal is to reduce that multi-prompt back and forth, which is very hard with PRDs but manageable if you do the upfront work.

I created skills in Claude that ask my team certain questions before we start writing: which customer are you solving for, which SKU do they belong to, which partner type? That way we’re prompted to supply all the relevant context before drafting, and there’s a template that structures the problem statement, use cases, and potential solutions. Once you give it that input, the quality of the output is significantly better. It takes a little longer to set up, but the result is worth it.

Improving AI Prompt Engineering and Output Quality

Rina Alexin | 00:11:52 – 00:13:20

You mentioned LLM as a judge. Can you walk through what that looks like in practice?

Advaita Nigudkar | 00:12:00 – 00:13:20

Once you have the structured PRD output, you use a second step where you feed the model a few of your best past PRDs and tell it: this is the gold standard. If the new PRD output does not meet this standard, flag what is missing and make sure the gaps are addressed. That’s what we compare it against to ensure consistency.

The reason this matters: teams are growing, the number of PMs is growing, and churn is high. How do you make sure that when a new PM joins, regardless of which product or team they’re on, there is a certain standard of quality and repeatability in what gets produced? The LLM as a judge helps us maintain that bar without relying on any one person’s head to carry it.

Using LLMs as Judges for Product Documentation

Rina Alexin | 00:13:20 – 00:15:39

Do you have the LLM as a judge just critique the PRD and have the PM own filling in the gaps? Or does the LLM also propose how to fix them?

Advaita Nigudkar | 00:14:26 – 00:15:39

The judge tells you the gaps. Then if you want, you can have the PM fill in the actual missing pieces, or you can ask the LLM to fill them. But if you just ask the LLM to fill it in without guidance, it goes back to the context problem: it doesn’t know your specific universe. The best practice is for the PM to give the inputs on what needs to go in, and then let the model finesse and polish it. The PM supplies the substance. The model handles the presentation.

Maintaining High Standards with AI-Generated Work

Rina Alexin | 00:15:39 – 00:17:28

I’ve been in situations where I encourage AI use and get output back that feels like it may have been reviewed, but sometimes these seven- to eight-page documents don’t hold up on close reading. Reading the same document for the fourth time, you get tired. How do you set expectations and maintain a high bar for what comes back from your team?

Advaita Nigudkar | 00:15:47 – 00:17:28

The team needs to spend more time figuring out what good looks like, and keep raising that bar. You’re right that documents are multiplying because they’re the easiest thing for an LLM to produce. But the accountability still sits with the PM. If you’re writing a PRD, you can absolutely use all the tools available, but the PRD still has to meet a certain standard so that engineering can take it and move with it.

The time LLMs are saving on the documentation itself needs to be reinvested into ensuring the quality is genuinely high before handing it to stakeholders. What I try to communicate is: spend more time talking to customers, more time providing higher-quality inputs, more time doing data analysis. You now have the means to improve. Use them. The expectation is not fewer hours of thinking. It’s better hours of thinking.

Enabling Teams with Standardized AI Prompts

Rina Alexin | 00:17:28 – 00:20:06

How much of your role as a product management leader involves making sure your team is enabled with standardized prompts so everyone is on the same page about what is expected at each stage?

Advaita Nigudkar | 00:18:44 – 00:20:06

This is a company-level effort, not just mine. Across product, engineering, and design, there’s work happening to standardize the entire product development lifecycle workflow and give teams the tools to do it consistently. There’s a subset of people building out skills that teams can invoke, whether for engineering, PM, or design work. The goal is that if you need to write a PRD, a business case, or a high-level design, the system already offers a structured way to do it out of the box, so teams are not reinventing the wheel every time.

Building AI Skills for Company-Wide Workflows

Rina Alexin | 00:20:06 – 00:21:09

Is product leadership involved in those decisions, or is it primarily an engineering-led initiative?

Advaita Nigudkar | 00:19:48 – 00:21:09

Yes. There are multiple tracks: one for engineering, one for product, one for design. Several product team members have been involved to make sure we’re building the right kinds of skills and agents to supercharge the work. For example, if you’re building a business case, someone has built a skill that gives you the entire template, prompts you for the right data, pulls in relevant information from integrations, and produces a draft. Proofreading and ensuring it meets the mark is still the PM’s responsibility, but the scaffolding is there.

Building Trust in AI in Regulated Environments

Rina Alexin | 00:21:09 – 00:23:02

You worked on launching an agentic AI workflow product at BILL, a regulated environment where money moves. How did that context shape the process? What did you have to think about that you wouldn’t in a less regulated space?

Advaita Nigudkar | 00:21:18 – 00:23:02

There were so many dimensions. On the information security side, when we pass data to an LLM we have to ensure it’s in encrypted format so we’re not sharing sensitive customer data with external tools. That required a lot of collaboration with InfoSec teams.

On the customer side, what our interviews surfaced was that trust is the central variable. We as PMs might feel confident that the feature is accurate 95% of the time, and mostly it is. But when it goes wrong, customers feel it immediately: how do I know I can trust this? What if you say a bill is $1,000 and it’s actually $100? How do I undo this? How do I know what AI has been doing in my account while I wasn’t there?

Those were the tension points we heard in research. To address them, we built auditability directly into the product: an activity log similar to an audit trail, showing exactly what the agent did, what decisions it made, and why. We also added controls so customers can turn individual AI capabilities on or off. If something goes wrong, they can go to that page, turn it off, and continue with confidence. That transparency is what made trust buildable.

Transparency Features for AI Products

Rina Alexin | 00:23:02 – 00:25:52

It sounds like you created a transparent way for users to understand what AI is doing in their account, because sometimes it can feel like too much magic. How important has this feature been to users? Is it heavily used?

Advaita Nigudkar | 00:24:31 – 00:25:52

It’s very important, but its importance is context-dependent. For lower-risk actions, like creating a bill that still goes through an approval chain before money moves, customers care less about the activity log. But for irreversible actions, it becomes critical.

The interesting part is that when everything is going well, customers do not need to see the activity log, and that is exactly how it should be. The engagement we initially saw dropped over time as customers built trust with the product. They checked it at first to understand what the agent was doing. Once they trusted it, they stopped checking unless something unexpected happened. A spike in activity log usage is not a good sign. It means something went wrong and customers needed to investigate.

Future Use Cases for AI in Product Management

Rina Alexin | 00:25:52 – 00:27:22

If AI continues to improve, what use case would be most valuable to you that isn’t possible today?

Advaita Nigudkar | 00:26:03 – 00:27:22

The piece I wish AI could do is work truly across teams. Agent-to-agent communication exists, but the negotiation layer, working with the context of PM versus engineering versus legal, risk, and compliance, is extremely complex. I would love to see AI get to a place where it can execute things end to end across those team boundaries without waiting on humans at every handoff.

This matters especially for global teams working across time zones. Teams in India, China, and the US waiting on each other is a real drag on speed. If AI could conduct some of those inter-team workflows without everyone needing to be online at the same moment, that would be genuinely transformative.

Gaining More Strategic Time with AI

Rina Alexin | 00:27:22 – 00:28:42

You’ve mentioned throughout our conversation that one of the biggest benefits of AI is that it frees up time. What do you do with that time?

Advaita Nigudkar | 00:27:45 – 00:28:42

I use it for more strategic work: customer interviews, deeper data analysis, understanding what we’re doing and how we’re doing it, reviewing work in progress, and doing product teardowns to ask what we can improve on what we’ve already shipped.

AI has opened up time for exactly the things product managers have always complained they didn’t have enough time to do. We spend so much time in meetings and in documentation. Now that AI is handling the rote parts, it creates the opportunity to go deeper in areas where we couldn’t before. That’s what’s going to help us build better products with fewer issues.

Advice for Product Managers on AI Adoption

Rina Alexin | 00:28:42 – 00:30:14

For a product manager trying to figure out where to incorporate AI and how seriously to take it, what’s your advice?

Advaita Nigudkar | 00:28:56 – 00:30:14

Everyone should take AI extremely seriously. There are so many benefits from just spending time learning about it, whether it’s different tools, MCPs, workflows. What you can gain professionally, and personally if you do it thoughtfully, is significant.

The main thing: leverage AI as an amazing tool and thought partner, but do not rely on it in a way that replaces the fundamental PM craft. Spend time early in your career, and throughout your career, building the foundational skills. Know your product inside and out. Then layer AI on top of that knowledge to do data work, research, prototyping, competitive teardowns. Imagine being the CEO of your product and using AI to go from initiation to launch, then doing the same thing with the next problem statement. You’ll learn an enormous amount.

The Importance of Curiosity in AI-Driven Product Management

Rina Alexin | 00:30:14 – 00:30:55

Is there a warning you’d give product managers as well?

Advaita Nigudkar | 00:30:29 – 00:30:55

Don’t over-rely on it in a way that costs you your curiosity. When you get everything easily, when you just ask and an output arrives, you stop being curious about the answer. You stop asking the next question. I find that dangerous.

Even for me, sometimes when I’m busy, I catch myself about to just ask AI something I should go investigate myself. That instinct to go look at a few customer recordings, to sit with a question, to go deeper than the first output, that’s what sets great PMs apart. Try not to let easy answers erode that quality in you.

Connecting with Advaita and Closing Remarks

Rina Alexin | 00:30:55 – 00:31:39

Advaita, thank you so much for joining me. How can our listeners follow you or connect with you?

Advaita Nigudkar | 00:31:10

LinkedIn is the best place. Please feel free to send me a request. It’s been a genuine pleasure talking with you, Rina.

Rina Alexin | 00:31:17 – 00:31:39

Thank you, and thank you all for listening to this episode of Productside Stories. If you liked today’s conversation, please don’t keep it to yourself. Share it with a friend and subscribe so you don’t miss a future episode. I’m Rina Alexin, and from all of us at Productside: let’s do product better, together.